Playbooks·03 Precision Sourcing
03 Precision Sourcing

Finding Deep-Tech
Talent

Deep-tech hiring follows different rules.

K

Kristina Golovko

MindDesign

7 min read

The deep-tech sourcing system

Specialisation

Define the exact technical domain

Signal

Output, not title or company

Ecosystem

Where this expertise concentrates

Sourcing approach

Precision, not volume

Deep-tech talent doesn't respond to generic outreach. It responds to evidence that you understand the work.

Why this matters

Deep-tech hiring fails when it uses standard recruiting logic.

Compiler engineers don't update LinkedIn profiles when they're open to opportunities. ML infrastructure specialists don't apply to job boards. Embedded systems experts are not reading recruiter InMails. The channels that work for general software hiring are largely invisible to the deep-tech talent pool.

Deep-tech candidates are often highly specialised, deeply embedded in specific communities, and primarily motivated by the quality of the technical problem — not compensation or brand. Reaching them requires understanding their world well enough to speak to it credibly.

The companies that hire exceptional deep-tech talent consistently share one characteristic: they find candidates where the work lives, not where recruiters look.

Founder reality

Before sourcing in any deep-tech domain, answer these:

01

Can we articulate the technical problem this role solves in terms that would resonate with an expert in this domain?

02

Do we know which conferences, communities, and institutions produce this expertise?

03

Are we searching for outputs — papers, contributions, systems — or just job titles?

04

How will we assess technical depth in evaluation? Is the interview process credible to an expert candidate?

05

What makes working on this problem intellectually interesting — and can we communicate that?

Deep-tech candidates evaluate the quality of the problem as carefully as the compensation. Sourcing must communicate both.

The framework

Four sourcing domains for deep-tech talent

Each deep-tech category has its own signal channels and community concentrations.

01

Compiler and systems engineering — LLVM, language design, low-level infrastructure

Find: LLVM Developers' Meeting and EuroLLVM speaker archives, LLVM GitHub contributors, academic compiler research groups (CMU, MIT, ETH Zurich), compiler-focused Discord and IRC communities, technical blog authors on systems topics.

02

ML infrastructure and research engineering — training systems, serving, MLOps

Find: PyTorch and JAX contributor lists, MLflow and Ray project contributors, MLOps community Slack channels, papers on ML systems at MLSys and NeurIPS, authors of technical posts on inference optimisation and distributed training.

03

Embedded and firmware engineering — IoT, real-time systems, hardware-adjacent

Find: RTOS community forums, Zephyr and FreeRTOS contributor lists, Hackaday and embedded.fm community, CCC (Chaos Communication Congress) speakers, hardware reverse engineering communities.

04

Cybersecurity and vulnerability research — offensive, defensive, firmware, protocol

Find: CVE author databases, DEF CON and Black Hat speaker archives, security-focused CTF team rosters, academic security research groups, Zero-day disclosure community participants.

Common mistakes

01

Using generic 'engineer' outreach for specialist roles

A compiler engineer who receives a message about 'exciting engineering opportunities' will not respond. The outreach must reference the specific technical domain and ideally the candidate's specific output.

02

Assuming the technical interview will validate depth

If sourcing produces candidates without genuine deep-tech background, the interview cannot recover the search. Depth must be pre-qualified through signal review before outreach.

03

Ignoring academic adjacent talent

PhD candidates and postdocs in relevant research areas are often the highest-potential source for deep-tech roles. Skipping academic communities eliminates a significant portion of qualified talent.

04

Moving too slowly once a strong candidate is engaged

Deep-tech candidates with genuine expertise are often in multiple conversations simultaneously. A slow process signals low conviction — and they move to companies that move faster.

Example scenario

A quantum computing startup, 25 people. Hiring a compiler engineer who specialises in quantum circuit optimisation. No prior hiring in this domain. Considered impossible internally.

The sourcing approach

Academic: reviewed PhD theses on quantum circuit optimisation from top 15 research universities. Identified 8 recent graduates or candidates.

Publications: searched arXiv for papers on quantum compilation in the past 3 years. Identified 12 authors with industry-relevant work.

Community: attended virtual IEEE Quantum Week — identified 4 speakers with directly relevant expertise.

LLVM adjacent: QIR (Quantum Intermediate Representation) GitHub contributors reviewed.

The outreach

Each message referenced the candidate's specific work — paper title, specific technical contribution, or talk topic. All mentioned the specific quantum compilation problem the role was hired to solve.

The outcome

19 targeted profiles. 11 messages sent. 9 responses. 3 candidates entered the process. Offer accepted from a recent PhD graduate — 6 months into a postdoc, actively seeking industry transition.

Deep-tech talent is not invisible — it's just not on the surface.

Finding it requires going where the work lives: publications, open-source, academic programs, specialist communities. The reward is access to candidates that volume sourcing will never reach.